Artificial intelligences are cross and delight of our times, advantages and disadvantages, opportunities and risks. Especially in the world of information and on social media, where it is increasingly complicated to distinguish the true from the false (here a recent example concerning Ukraine).
Also in view of the many elections scheduled for 2024, from the European elections to the American presidential elections, more or less all platforms are working to counter the spread of so-called synthetic media (i.e. content created or modified with AI) or at least to label them, so that those who see them can immediately know what is in front of them. Among those most involved in this field is undoubtedly TikTok: not only has it had a series of rules for artificial videos (these) and asks creators to indicate them as such before uploading, but now it is also the first and only video platform to use artificial intelligence to identify them. It uses an AI to recognize other AIs, which it has done for years seems to be the only possible solution to this type of problem.
In-depth information Accusation, defense, timing, how it could end: everything you need to know about Emanuele Capone’s TikTok affair 06 May 2024
Adobe’s filter when uploading videos
The chosen tool is somehow (almost) as important as the result, because it is a bit the industry standard: how it should do also Instagram with imagesTikTok will use the AI created by Adobe for the Content Credentials projectwhich is used precisely to identify and label AIGC, i.e. AI-generated content.
The novelty lies precisely in the fact that this thing happens per i video and that it happens not only for those created with the tools inside the TikTok app but even with those made elsewhere. For example, if you create with CapCut (or with one of the software produced by Adobe or OpenAI) a clip starting from a deepfake, exporting it to your smartphone and then uploading it to TikTok as if it were real content, the platform will recognize it as being made with AI. Adam Presser, who is responsible for the Operations & Trust and Safety division at TikTok, confirmed that “this allows us to extend automatic labeling to AIGCs created on some other platformsa capability that we plan to progressively roll out to all users globally in the coming weeks.”
In short, TikTok has established an important one partnership con la Coalition for Content Provenance and Authenticity (abbreviated as C2PA), a group in which the leading companies in the tech world have gathered precisely to establish standards for the correct use of AI: “In the next few months we will also begin to attach Content Credentials to TikTok contents, in which they will remain even once downloaded – added Presser – This means that anyone will be able to use the C2PA verification tool to identify AIGCs made on TikTok and even to know when, where and how the content was created or modified. In turn, other platforms that adopt Content Credentials will be able to automatically label it.”
This thing is possible because Adobe’s AI works as a sort of filter during the upload phase: it checks the video and if it recognizes some intervention by another AI, it attaches information to the content that the TikTok software can recognize and use to apply the AI Generated label (image above). The fact that this marking occurs at the metadata level means that it is not (or should not be) removable even in the event of download and reuse.
tiktok: an example of MediaWise content
ChatGPT, Dall-E and fake news: this is how OpenAI is preparing for the elections by Bruno Ruffilli 17 January 2024
Collaboration with MediaWise to help minors
From TikTok they confirmed to us that, “to help promote the adoption of Content Credentials in our sector, we are also joining the Content Authenticity Initiative”, again led by Adobe: “We are very pleased to be able to welcome TikTok to both C2PA and CAI, helping to support the journey of its large global community of creators and users towards ever greater transparency and authenticity – commented Dana Rao, CTO of Adobe – In an era where any digital content can be altered, it is essential to provide the public the possibility of understanding what is real. And this announcement is a fundamental step towards this result.”
That’s not all, because TikTok has also anticipated the arrival of new features dedicated to the so-called media literacy, specifically applied to AI and developed together with MediaWise, an interesting platform (active online since 2018) dedicated to combating misinformation and aimed above all at teenagers, who are those who can most easily fall into the traps of synthetic media: “Since 2019, with our Teen Fact-Checking Network we train the public on TikTok with media literacy videos – recalled Alex Mahadevan, director of MediaWise – After 5 years, we are happy to offer an even larger number of people the tools to separate fact from fiction”.
Obviously, the one adopted by TikTok it is not the solution to all problems and it would be naive to expect that there will be no more AI deceptions, that some of this content will slip through the cracks, and that there won’t be more people being deceived. But it is a significant step in the right direction, especially in this election year. It will become even more so as more and more platforms use Content Credentials, as they will be more and more recognizable and labelable contents come AI Generated.
@capoema